System Identification of Building from Strong Motions

System Identification of Building from Strong Motions

Recursive parameter estimation method is used for on-line system identification. It is called RARX model. In this model, an initial covariance matrix should be assigned first but the covariance matrix may highly affect the identification results. Therefore, in this paper, artificial neural network and RARX model are combined. The weightings in the single-layer artificial neural network are taken as the parameters to be estimated. It is called RARX-ANN model. No initial guess for the covariance matrix is necessary. The non-linear and time-varying system of the Taitung Fire Company Building during the Taitung earthquake on April 1, 2006 was identified by the model. From the results of identification, the frequency of the building was decreased and the damping ratio was increased in the earthquake due to non-linear behavior induced by damage.